Agentic security survey finds containment lags runtime permissions
A July survey found 65% of active enterprise agent programs enforce scoped permissions, while only 18% isolate high-risk agents.
By Wei-Lin Zhao 1 week agoSection
AI
Agent evaluation reliability survey finds autonomy gap after customer failures
A July survey found firms reporting eval-passing AI failures were more likely to pursue zero-human deployment, though it cannot ex...
By Colin Brandt 1 week agoEnterprise agent cost controls remain post-hoc for 21% in Pulse survey
A July survey of 107 organizations found broad investment in agent oversight, while 21% reported no real-time way to halt runaway...
By Wei-Lin Zhao 1 week agoGoverned AI context layers expose more recurring agent errors, survey finds
A July survey found a 50% recurring-error reporting rate among enterprises building context layers, versus 21% without them.
By Colin Brandt 1 week agoEnterprise AI compute costs trail production deployment, survey finds
A July survey of 170 enterprises found 66% run AI in production, while 47% rigorously track compute cost and return.
By Renata Fuchs 1 week ago
Mistral EU data processing adds regional controls, with limits
Mistral’s EU and US inference endpoints add location controls, but exclude several stateful features and carry a 10% premium.
By Wei-Lin Zhao 1 week ago
ChatGPT Business Premium seats cost $125 a month for added capacity
OpenAI’s new ChatGPT Business Premium seats offer five times Standard usage and remove the five-hour limit, with a waitlist ahead...
By Renata Fuchs 1 week agoOpenAI GPT-5.6-Cyber targets vetted defenders with fewer cyber refusals
OpenAI’s GPT-5.6-Cyber reported a 95% response rate on its internal dual-use cyber test, but the figure is not an accuracy score.
By Colin Brandt 1 week ago
Anthropic Claude watermarks roll out globally on supported new models
Anthropic will mark output from supported new Claude models worldwide, but a watermark signals processing, not proof of authorship...
By Renata Fuchs 1 week ago
OpenAI GPT-5.6-Cyber launches through restricted Daybreak Red access
OpenAI’s cyber-specific GPT-5.6 model cuts refusals for approved researchers, but its benchmark lead over Sol is uneven.
By Renata Fuchs 1 week ago
FineBooks OCR benchmark puts a price on cleaning historical books for AI
FineBooks tested 14 open OCR models on historical books, finding 97% accuracy is feasible for AI training but not scholarly use.
By Colin Brandt 1 week agoMeta Muse Glimmer release puts 30B agent model under Apache 2.0
Meta has released Muse Glimmer’s weights under Apache 2.0, targeting local AI agents on high-end consumer hardware.
By Colin Brandt 1 week ago